Comparison Between Feature Extraction Algorithms for Sentiment Recognition from Text

The amount of text being produced nowadays is growing dramatically daily. Social media networks like Instagram and Twitter, in-app messaging services, e-commerce websites, Google searches, news websites, and a wide range of additional sources are just a few examples of potential providers. These sources collectively provide gigantic amounts of textual data every single instant. The term “Text Sentiment Analysis” (TSA) represents an ensemble of methodologies that can be used to extract the key characteristics from textual data and then categorize them to identify the precise sentiments. Pre-processing, feature extraction, and sentiment categorization are all parts of the sentiment analysis process as the text data we get from these media is unstructured by nature. However, the selection of the features in this scenario has a significant impact on detecting the polarity of text. In this work, three fundamental feature extraction techniques—TF-IDF, BOW, and N-gram are compared for textual sentiment recognition using the balanced dataset collected from tweets i.e., Sentiment140. This study serves as the foundation for comprehending the domain's complexity and the level of efficiency attained by the various approaches. The results of the evaluation show that classification performance is significantly influenced by the preprocessing strategy, feature extraction methods.